Efficient stochastic sensitivity analysis of discrete event systems

Sensitivity analysis quantifies the dependence of a system’s behavior on the parameters that could possibly affect the dynamics. Calculation of sensitivities of stochastic chemical systems using Kinetic Monte Carlo and finite-difference-based methods is not only computationally intensive, but direct calculation of sensitivities by finite-difference-based methods of parameter perturbations converges very poorly. In this paper we develop an approach to this issue using a method based on the Girsanov measure transformation for jump processes to smooth the estimate of the sensitivity coefficients and make this estimation more accurate. We demonstrate the method with simple examples and discuss its appropriate use.